Live Platform Data
Drux Activity
Real-world AI model performance from every query run on Drux. Speed, reliability, and consensus measured on actual user questions — not controlled benchmarks.
647 searches · 1,992 model calls · updated every 5 min
Platform Summary
647 total searches completed. 1992 individual model calls recorded. Average consensus score: 61%. Model completion rate: 73%.
Top Models by Speed
- inception/mercury-2: avg 3.9s, 98% reliability, 110 queries
- google/gemini-2.5-flash: avg 7.0s, 40% reliability, 5 queries
- mistralai/mistral-small-2603: avg 8.6s, 100% reliability, 136 queries
- cohere/command-r-08-2024: avg 8.7s, 89% reliability, 81 queries
- x-ai/grok-4.3: avg 8.7s, 100% reliability, 11 queries
Most Agreed Questions
- BeEzrat HaShem Inc. Earns Candid Platinum Seal of Transparency for the 3rd Time — consensus score 100%
- What if users start cloning SaaS using AI — consensus score 90%
- Why about a third of the submissions become dead in mere minutes? — consensus score 90%
- Is HN crowd a left-leaning? — consensus score 90%
- How Do You Connect OpenAI Secure MCP Tunnel with Claude Desktop — consensus score 90%
Platform Pulse
Total Searches
647
This Week
77
last 7 days
Today
0
Model Calls
1,992
across all searches
Avg Consensus Score
61%
across all completed searches
Model Completion Rate
73%
calls that returned successfully
Model Leaderboard — on Drux
Last 647 searches · click column to sort
| Model | Avg Speed▲ |
|---|---|
1Mercury 2FREE | 3.9s |
2Gemini 2.5 FlashFREE | 7.0s |
3Mistral Small 4FREE | 8.6s |
4Command RFREE | 8.7s |
5Grok 4.3PAID | 8.7s |
6Phi-4FREE | 8.7s |
7Hermes 3 70BFREE | 12.4s |
8GPT OSS 120BFREE | 12.9s |
9Llama 4 MaverickFREE | 13.2s |
10Claude Sonnet 4.6PAID | 14.1s |
11Qwen3 235BFREE | 14.2s |
12Gemini 3.5 FlashPAID | 14.3s |
13DeepSeek V3FREE | 17.0s |
14Sonar ProPAID | 17.8s |
15Gemma 3 27BFREE | 19.3s |
16Solar Pro 3FREE | 20.3s |
17Seed 1.6FREE | 23.7s |
18GPT-5.5PAID | 24.6s |
19Nemotron 49BFREE | 24.8s |
20GPT-5 MiniFREE | 25.3s |
21ERNIE 4.5 300BFREE | 27.8s |
22Claude Opus 4.7PAID | 31.1s |
23OLMo 3 32BFREE | — |
⚡ Fastest on Drux
Mercury 2
avg 3.9s · 98% reliable
✓ Most Reliable on Drux
Mistral Small 4
100% completion · 136 calls
Consensus Digest — Recent Public Searches
✓ Models Agreed
BeEzrat HaShem Inc. Earns Candid Platinum Seal of Transparency for the 3rd Time
What if users start cloning SaaS using AI
Why about a third of the submissions become dead in mere minutes?
Is HN crowd a left-leaning?
How Do You Connect OpenAI Secure MCP Tunnel with Claude Desktop
⚡ Models Diverged
How many failed startups have you launched?
What are you building first with Fable?
Got access to Gemini's actual thinking
The biggest issue is the final paragraph: “The portals are open. The money is there. No connections needed.” This cannot be verified as a blanket statement. * Some programmes have application windows, not permanently open portals. * Some are cohort-based or depend on funding availability. * Meeting eligibility requirements does not guarantee acceptance or funding. * Demand often exceeds available spaces. Overall assessment * About 80–90% accurate. * Real government programmes: ✔️ * Some benefits simplified or exaggerated: ✔️ * “All portals are open” and “money is there” are not verifiable universal claims: ✔️ If you were sharing this publicly, a more accurate ending would be: “These are legitimate government programmes. Availability, eligibility, and application windows vary, so check each official portal for current application status and requirements before applying.” That version is much easier to verify and avoids implying that every programme is currently accepting applications or has guaranteed funding.
Who has been caught out by this anti-pattern from Google?
Updated every 5 min · Data from last 1,000 searches
OpenRouter Usage
Raw model traffic from Drux's OpenRouter account. Tokens in = prompts we send; tokens out = model responses.
Lifetime key spend: $93.87 · data refreshes hourly from OpenRouter
About This Data
How is this different from other AI benchmarks?
Most AI benchmarks are run under controlled laboratory conditions on standardised test sets. Drux Activity measures performance on real user questions — diverse, unpredictable, and representative of actual use. Speed and reliability numbers here reflect what users actually experience.
What is a Consensus Win?
When multiple models answer the same question and their responses agree closely (consensus score ≥ 7/10), all models that responded are credited with a Consensus Win. A high Consensus Win rate means a model consistently lands on the same answer as its peers — a signal of reliability beyond just accuracy.
Why do free models have more queries?
Drux randomly selects models for each search. Free-tier models are included in more searches because they are available to all users. Paid models are selected when users opt into premium tiers. Query count reflects availability and tier distribution, not quality — sort by Speed or Reliability for a fairer comparison.
How often is this updated?
The leaderboard updates every 5 minutes, pulling from the last 1,000 completed searches. As query volume grows the data becomes more statistically significant.